BICLIC

BICLIC: Biclustering for the comprehensive search of correlated gene expression patterns using clustered seed expansion

Introduction:

In this paper an algorithm is proposed for biclustering analysis. The proposed method is named as BIclustering by Correlated and Large number of Individual Clustered seeds (BICLIC). Unlike other past algorithm that depends on random seeds to start, the BICLIC perform the biclustering by using the individually dimensioned clusters.

Features:

• Biclustering algorithm
• Utilization of individually clustered sets of gene
• Full microarray data search for Biclusters
• Stable output
• High accuracy
• Functional enrichment evaluation of obtained clusters using COFECO

Working:

The BICLIC use the initial seeds obtained from individual dimension clustering. For individual dimension clustering it uses the same concept as proposed in CLIC but for both gene and conditions. After getting the initial seed that are found from searching the overall dataset, the BICLIC expand the search for other bi-clusters. As shown in figure below it uses the Pearson correlation coefficient (PCC) to measure the similarity between the seed cluster and newly find clusters. Those clusters that are correlated with seed clusters but contain different genes will be considered as actually clusters. To expand the search for new cluster the BICLIC increase the row and columns in such a way that when PCC value maintain with in the defined threshold. Once the column expansion is done the same rule used in the expansion of rows. Then the rectangular region of is used to search the clusters and now reduction of rows and columns is performed till the threshold is met.

Performance:

• The performance evaluation BICLIC is compared with
  • BCCA
  • CPB
  • QUBIC
• Functional enrichment evaluation of obtained clusters using COFECO
• Real and simulated data for evaluation
  • simulated data with following patterns: shifting, scaling, and shifting-scaling
  • real yeast and lung cancer microarray data
• Following performance measures were used to evaluate the overall results
  • Cluster count
  • Average cluster size
  • Gene coverage
  • Condition coverage
  • Cell coverage
  • Average relevancy of obtained clusters with simulated conditions
  • Average recovery vs column fraction level
  • Number of enriched terms
  • Proportion of clusters after removal of overlaps

Conclusion:

BICLIC is a biclustering algorithm, it utilizes the seeds obtained from the individual dimension clustering. The algorithm was implemented on R, and it is compared with three different biclustering algorithms on various performance measures.

Example

You can download the package from http://bisyn.kaist.ac.kr/software/BICLIC_1.0.tar.gz

1. Install

install.packages("http://bisyn.kaist.ac.kr/software/BICLIC_1.0.tar.gz", repos=NULL, type="source")

## Installing package into 'C:/Users/sypark/Documents/R/win-library/3.3'
## (as 'lib' is unspecified)

There are 18 functions included in the package BICLIC

library(BICLIC)

2. Working Example (Underconstruction)

plot(pressure)

Possible Improvements:

• By utilizing better clustering algorithm for individual dimension cluster
• Online web-based tool development for biclustering with built-in features for functional enrichment and the selection of parameters like number of clusters, size of clusters and clustering technique selection.

Comparison of Biclustering Algorithms:

A comparative analysis of biclustering algorithms for gene expression data, Kemal Eren et. al., Briefings in Bioinformatics, 2012 http://bib.oxfordjournals.org/content/early/2012/07/06/bib.bbs032

A comparison and evaluation of five biclustering algorithms by quantifying goodness of biclusters for gene expression data, Li Li et. al., Bio data Mining, 2012 https://biodatamining.biomedcentral.com/articles/10.1186/1756-0381-5-8

A comparative study of clustering and biclustering of microarray data, Haifa Ben Saber et. al., IJCSIT, 2014 https://doaj.org/article/d7b38ce8103c45c69b624316cf550177

Biclustering on expression data: A review, Beatriz Pontes et. al., JBI, 2015 http://dx.doi.org/10.1016/j.jbi.2015.06.028